Electronic component manufacturing system based on information perception and intelligent optimization

Abstract

The invention belongs to the technical field of information, and particularly focuses on an electronic component manufacturing system based on information perception and intelligent optimization, after a multi-sensor fusion and real-time correction algorithm is adopted, the surface mounting state can be comprehensively perceived, the path is optimized through an advanced fusion theory and a filtering algorithm, the rejection rate is greatly reduced, the chip position deviation is well controlled, and the product quality is improved. And the production efficiency is obviously improved. When a computer mainboard is welded, traditional fixed parameter welding cannot cope with differences of materials and the like, and welding defects are many. Multi-physical field monitoring and self-adaptive control algorithms are applied, multi-sensor data are fused, power is adjusted through fuzzy control after principal component analysis and dimension reduction, welding defects are effectively reduced, the strength consistency of welding spots is remarkably improved, and the welding quality and reliability are improved.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and in particular relates to an electronic component manufacturing system based on information perception and intelligent optimization. Background Art

[0002] The electronic component manufacturing industry, as the foundation of the modern electronic information industry, has a direct impact on the competitiveness of the entire sector. With the rapid advancement of electronic technology, electronic components are becoming smaller and more complex, placing increasing demands on manufacturing precision and quality. Industrial robots, with their high precision, high speed, and high stability, have been widely used in various aspects of electronic component manufacturing, such as chip placement, circuit board soldering, and component testing. However, existing industrial robot technology still faces numerous challenges in the electronic component manufacturing process, severely restricting the further development of the industry.

[0003] Chip placement accuracy is difficult to guarantee: Chip placement is a critical step in electronic component manufacturing, and placement accuracy directly impacts the performance and reliability of electronic components. Traditional industrial robotic chip placement systems rely primarily on visual positioning and fixed placement parameters, making them unable to accurately adapt to factors such as subtle variations in chip size, variations in circuit board flatness, and vibration during the placement process. Furthermore, nozzle wear and chip offset can easily lead to significant deviations in chip placement position, impacting the proper functioning of electronic components, increasing scrap rates, and reducing production efficiency.

[0004] Unstable PCB soldering quality: PCB soldering is a critical process for connecting electronic components to circuit boards. The quality of soldering is directly related to the electrical performance and service life of these components. Existing industrial robotic PCB soldering systems primarily utilize preset soldering parameters and fixed soldering paths, which are unable to adjust in real time based on the PCB material, thickness, component layout, or the actual soldering process. Furthermore, due to issues such as uneven soldering temperatures and inaccurate soldering volume control during the soldering process, soldering defects such as cold solder joints, short circuits, and incomplete solder joints can easily occur, impacting the soldering quality and reliability of the PCB.

[0005] Low electronic component inspection efficiency and accuracy: Electronic component inspection is a crucial means of ensuring product quality, and its efficiency and accuracy directly impact production cycles and product quality. Currently, industrial robots primarily rely on manual spot checks and simple automated testing equipment for electronic component inspection. This approach is inefficient, subjective, and prone to missed detections, making it unable to meet the inspection needs of large-scale electronic component production. Furthermore, existing inspection algorithms are unable to accurately identify and classify complex electronic component defects, making it difficult to locate and quantify defects in real time. This hinders the timely detection and resolution of electronic component quality issues. Summary of the Invention

[0006] The present invention provides an electronic component manufacturing system based on information perception and intelligent optimization, comprising:

[0007] The chip high-precision placement algorithm module based on multi-sensor fusion and real-time correction obtains multi-source information in the chip placement process by installing multiple sensors on the placement head of the industrial robot. It uses multi-sensor fusion technology and real-time correction algorithm to achieve high-precision chip placement. Let the chip position data detected by the visual sensor be Pv, the placement pressure data detected by the pressure sensor be Fp, the nozzle displacement data detected by the displacement sensor be Ds, and the circuit board flatness data detected by the laser ranging sensor be Hl. After fusion using the Dempster-Shafer evidence theory, the fused data M is obtained; for the evidence source Si corresponding to each sensor data, its basic probability distribution function is mi, and through orthogonal sum operation,

[0008] m=m1⊕m2⊕…⊕mn to obtain the basic probability distribution m after fusion, and then determine the fusion data M; based on the real-time correction algorithm, Kalman filtering is used to optimize the placement path. Let the state transfer matrix be A and the observation matrix be H

[0009] , the process noise covariance is Q, the observation noise covariance is R, the initial state estimate is x^0|0, the covariance estimate is P0|0, at time k, first predict: x^k|k-1=Ax^k-1|k-1, Pk|k-1=APk-1|k-1AT+Q; then update: Kk=Pk|k-1HT(HPk|k-1HT+R)-1,

[0010] x^k|k=x^k|k-1+Kk(zk-Hx^k|k-1), Pk|k=(I-KkH)Pk|k-1, where

[0011] zk is the observation value corresponding to the fused data M. The above-mentioned Kalman filtering process is used to adjust the placement path and other parameters in real time to ensure the chip placement accuracy;

[0012] The circuit board welding quality optimization algorithm module based on multi-physics field monitoring and adaptive control installs multiple sensors on the welding equipment of the industrial robot to obtain the multi-physics field parameters during the welding process. It uses multi-source information fusion technology and adaptive control algorithms to optimize the circuit board welding quality. Let the welding temperature data detected by the temperature sensor be T, the welding current data detected by the current sensor be I, the solder joint shape data detected by the visual sensor be S, and the internal structure data of the welding part detected by the ultrasonic sensor be V. After dimensionality reduction and fusion through principal component analysis (PCA), the comprehensive data N is obtained. First, the original data matrix X = [T, I, S, V] T is centered and the covariance matrix C = n-11XXT is calculated.

[0013] Then, the eigenvalues ​​λi and eigenvectors ei of the covariance matrix are solved, and the first m principal components with a cumulative contribution rate of more than 85% are selected to obtain the transformation matrix W = [e1, e2, ..., em], the comprehensive data N = WTX, and the welding power is adjusted by fuzzy control based on the adaptive control algorithm. Assume that the input of the fuzzy controller is the comprehensive data N and the expected deviation e of the welding quality and its rate of change Δe, and the output is the welding power adjustment ΔPw. The fuzzy set of input and output variables is defined. The fuzzy set of e is {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}. Through the fuzzy rule base, such as "If e is positive large and Δe is positive small, then ΔPw is positive large", the welding power adjustment ΔPw is obtained through fuzzy reasoning and defuzzification process. By adjusting the welding parameters in real time, the welding quality is optimized;

[0014] The electronic component rapid detection algorithm module based on multimodal data fusion and deep learning achieves rapid detection of electronic components by installing multiple detection devices on the electronic component detection production line to obtain multimodal data of electronic components and using multimodal data fusion technology and deep learning algorithms;

[0015] The reinforcement learning-based multi-robot collaborative electronic component manufacturing scheduling algorithm module treats multiple industrial robots in the electronic component manufacturing workshop as a multi-agent system, defines the state space, action space, and reward function, and uses the reinforcement learning algorithm to achieve efficient collaborative work among multiple robots;

[0016] The electronic component manufacturing process optimization algorithm module based on digital twin and virtual simulation establishes a digital twin model of the electronic component manufacturing system and uses virtual simulation to simulate and optimize the manufacturing process of electronic components, thereby achieving optimization and improvement of the electronic component manufacturing process.

[0017] Furthermore, in the high-precision chip placement algorithm module based on multi-sensor fusion and real-time correction, the accuracy of the visual sensor is ±0.01mm, the accuracy of the pressure sensor is ±0.05N, the accuracy of the displacement sensor is ±0.02mm, and the accuracy of the laser ranging sensor is ±0.03mm. The sensor deployment position is determined by combining finite element simulation and actual testing to ensure accurate data collection. The sensor acquisition frequency is 10Hz in the placement preparation stage, and is increased to 50Hz in the placement execution stage, and is dynamically adjusted according to the placement process.

[0018] Furthermore, in the circuit board welding quality optimization algorithm module based on multi-physical field monitoring and adaptive control, the temperature sensor accuracy is ±1°C, the current sensor accuracy is ±0.1A, the visual sensor accuracy is ±0.02mm, and the ultrasonic sensor resolution is ±0.05mm. The sensors work together to provide high-frequency and precise data for the adaptive control algorithm, and data transmission adopts a low-latency, high-bandwidth industrial Ethernet protocol.

[0019] Furthermore, in the electronic component rapid detection algorithm module based on multimodal data fusion and deep learning, the resolution of the visual inspection equipment is 0.01mm, the penetration depth resolution of the X-ray inspection equipment is 0.05mm, and the temperature resolution of the infrared detection equipment is ±0.2°C. Multimodal data fusion adopts a fusion network based on deep learning. The network structure contains 3 convolution layers, and the convolution kernel sizes are 3×3, 5×5, and 3×3 respectively. The network is trained with a large amount of electronic component defect data (more than 8,000 groups) and the number of training times is 2,500 to improve the defect recognition ability.

[0020] Furthermore, in the multi-robot collaborative electronic component manufacturing scheduling algorithm module based on reinforcement learning, the communication delay between intelligent agents is less than 10ms, the communication reliability reaches 99.9%, and real-time interaction is achieved through a distributed communication network. When training the reinforcement learning model, the experience replay buffer capacity is set to 15,000 data, and the target network update frequency is once every 120 iterations.

[0021] Furthermore, in the electronic component manufacturing process optimization algorithm module based on digital twins and virtual simulation, professional modeling software such as SolidWorks is used to build a digital twin model. The model update frequency is consistent with the sensor data acquisition frequency to ensure real-time synchronization between the physical entity and the virtual model. By simulating the chip mounting stress distribution and circuit board welding temperature field under different process parameters, potential problems are predicted in advance, and Monte Carlo simulation is used to perform uncertainty analysis on the prediction results. The number of simulations is set to 1200 times.

[0022] Furthermore, in the high-precision chip placement algorithm module based on multi-sensor fusion and real-time correction, the Dempster-Shafer evidence theory uses an improved discount coefficient method to process conflicting evidence during fusion to improve fusion accuracy. During the Kalman filtering process, the state transition matrix A and the observation matrix H are determined based on the chip placement dynamics model and sensor measurement characteristics. The process noise covariance Q and the observation noise covariance R are obtained through statistical analysis of experimental data.

[0023] Furthermore, in the circuit board welding quality optimization algorithm module based on multi-physical field monitoring and adaptive control, when principal component analysis (PCA) is used for dimensionality reduction, the number of principal components is determined by calculating the eigenvalue contribution rate, the fuzzy control rule base is established based on the experience of welding process experts and experimental data, and the fuzzy reasoning adopts the Mamdani reasoning method, and the defuzzification adopts the center of gravity method.

[0024] Furthermore, in the electronic component rapid detection algorithm module based on multimodal data fusion and deep learning, the deep learning model adopts a structure that combines convolutional neural network (CNN) and recurrent neural network (RNN). CNN extracts the spatial features of electronic components, and RNN processes time series features. The features are further processed through a bidirectional long short-term memory network (Bi-LSTM) to improve the accuracy of defect identification and quantitative analysis.

[0025] Furthermore, in the multi-robot collaborative electronic component manufacturing scheduling algorithm module based on reinforcement learning, the reward function is based on the manufacturing efficiency improvement ratio α, product quality improvement ratio β, and equipment utilization improvement ratio γ.

[0026] To construct a quantitative indicator, the reward value R = w1α + w2β + w3γ, and w1 = 0.4 is determined by the hierarchical analysis method.

[0027] , w2=0.3, w3=0.3, to guide the robot to learn the optimal collaborative manufacturing scheduling strategy.

[0028] Beneficial effects:

[0029] Improve assembly quality inspection accuracy: The high-precision assembly quality inspection algorithm based on multimodal sensor fusion can comprehensively and accurately obtain quality information during the assembly process, achieve high-precision inspection of assembly quality, promptly discover and correct minor quality problems, and reduce product defective rates.

[0030] Achieve full-process quality traceability: The full-process quality traceability system based on blockchain technology ensures the integrity, accuracy and non-tamperability of quality traceability information, can quickly and accurately locate the root cause of quality problems, and provide strong support for quality improvement.

[0031] Enhanced real-time and flexible quality control: Adaptive quality control algorithms based on real-time data analysis monitor the status of the assembly process and quality trends in real time, proactively adjusting the robot's assembly parameters and operating strategies to achieve real-time control of assembly quality. Furthermore, the system can automatically trigger emergency response mechanisms based on actual conditions, enhancing quality control flexibility.

[0032] Optimize assembly process and quality: The virtual assembly and quality optimization system based on digital twins simulates and emulates the assembly process in a virtual environment, predicts possible quality problems in advance, and proposes corresponding improvement measures to achieve continuous optimization of assembly process and quality.

[0033] Improve multi-robot collaborative efficiency: The multi-robot collaborative quality control and intelligent scheduling system realizes information sharing and collaborative operation among multiple robots, rationally allocates tasks and schedules resources, and improves the overall quality and efficiency of the assembly line. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Module flow diagram. DETAILED DESCRIPTION

[0035] Example 1

[0036] An electronic component manufacturing system based on information perception and intelligent optimization, comprising:

[0037] The chip high-precision placement algorithm module based on multi-sensor fusion and real-time correction obtains multi-source information in the chip placement process by installing multiple sensors on the placement head of the industrial robot. It uses multi-sensor fusion technology and real-time correction algorithm to achieve high-precision chip placement. Let the chip position data detected by the visual sensor be Pv, the placement pressure data detected by the pressure sensor be Fp, the nozzle displacement data detected by the displacement sensor be Ds, and the circuit board flatness data detected by the laser ranging sensor be Hl. After fusion using the Dempster-Shafer evidence theory, the fused data M is obtained; for the evidence source Si corresponding to each sensor data, its basic probability distribution function is mi, and through orthogonal sum operation,

[0038] m=m1⊕m2⊕…⊕mn to obtain the basic probability distribution m after fusion, and then determine the fusion data M; based on the real-time correction algorithm, Kalman filtering is used to optimize the placement path. Let the state transfer matrix be A and the observation matrix be H

[0039] , the process noise covariance is Q, the observation noise covariance is R, the initial state estimate is x^0|0, the covariance estimate is P0|0, at time k, first predict: x^k|k-1=Ax^k-1|k-1, Pk|k-1=APk-1|k-1AT+Q; then update: Kk=Pk|k-1HT(HPk|k-1HT+R)-1,

[0040] x^k|k=x^k|k-1+Kk(zk-Hx^k|k-1), Pk|k=(I-KkH)Pk|k-1, where

[0041] zk is the observation value corresponding to the fused data M. The above-mentioned Kalman filtering process is used to adjust the placement path and other parameters in real time to ensure the chip placement accuracy;

[0042] The circuit board welding quality optimization algorithm module based on multi-physics field monitoring and adaptive control installs multiple sensors on the welding equipment of the industrial robot to obtain the multi-physics field parameters during the welding process. It uses multi-source information fusion technology and adaptive control algorithms to optimize the circuit board welding quality. Let the welding temperature data detected by the temperature sensor be T, the welding current data detected by the current sensor be I, the solder joint shape data detected by the visual sensor be S, and the internal structure data of the welding part detected by the ultrasonic sensor be V. After dimensionality reduction and fusion through principal component analysis (PCA), the comprehensive data N is obtained. First, the original data matrix X = [T, I, S, V] T is centered and the covariance matrix C = n-11XXT is calculated.

[0043] Then, the eigenvalues ​​λi and eigenvectors ei of the covariance matrix are solved, and the first m principal components with a cumulative contribution rate of more than 85% are selected to obtain the transformation matrix W = [e1, e2, ..., em], the comprehensive data N = WTX, and the welding power is adjusted by fuzzy control based on the adaptive control algorithm. Assume that the input of the fuzzy controller is the comprehensive data N and the expected deviation e of the welding quality and its rate of change Δe, and the output is the welding power adjustment ΔPw. The fuzzy set of input and output variables is defined. The fuzzy set of e is {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}. Through the fuzzy rule base, such as "If e is positive large and Δe is positive small, then ΔPw is positive large", the welding power adjustment ΔPw is obtained through fuzzy reasoning and defuzzification process. By adjusting the welding parameters in real time, the welding quality is optimized;

[0044] The electronic component rapid detection algorithm module based on multimodal data fusion and deep learning achieves rapid detection of electronic components by installing multiple detection devices on the electronic component detection production line to obtain multimodal data of electronic components and using multimodal data fusion technology and deep learning algorithms;

[0045] The reinforcement learning-based multi-robot collaborative electronic component manufacturing scheduling algorithm module treats multiple industrial robots in the electronic component manufacturing workshop as a multi-agent system, defines the state space, action space, and reward function, and uses the reinforcement learning algorithm to achieve efficient collaborative work among multiple robots;

[0046] The electronic component manufacturing process optimization algorithm module, based on digital twins and virtual simulation, simulates and optimizes the manufacturing process of electronic components by establishing a digital twin model of the electronic component manufacturing system and utilizing virtual simulation to achieve optimization and improvement of the electronic component manufacturing process. In the high-precision chip placement algorithm module based on multi-sensor fusion and real-time correction, the visual sensor accuracy is ±0.01mm, the pressure sensor accuracy is ±0.05N, the displacement sensor accuracy is ±0.02mm, and the laser ranging sensor accuracy is ±0.03mm. The sensor deployment positions are determined through a combination of finite element simulation and actual testing to ensure accurate data collection. The sensor acquisition frequency is 10Hz during the placement preparation phase, increasing to 50Hz during the placement execution phase, and dynamically adjusted according to the placement process. In the circuit board welding quality optimization algorithm module based on multi-physics field monitoring and adaptive control, the temperature sensor accuracy is ±1°C, the current sensor accuracy is ±0.1A, the visual sensor accuracy is ±0.02mm, and the ultrasonic sensor resolution is ±0.05mm. Each sensor works together to provide high-frequency and accurate data for the adaptive control algorithm, and the data transmission adopts the low-latency, high-bandwidth industrial Ethernet protocol. In the electronic component rapid detection algorithm module based on multimodal data fusion and deep learning, the visual detection equipment resolution is 0.01mm, the X-ray detection equipment penetration depth resolution is 0.05mm, and the infrared detection equipment temperature resolution is

[0047] ±0.2°C. Multimodal data fusion utilizes a deep learning-based fusion network. The network structure consists of three convolutional layers with kernel sizes of 3×3, 5×5, and 3×3, respectively. The network is trained 2500 times on a large amount of electronic component defect data (over 8,000 sets) to improve defect recognition capabilities. In the reinforcement learning-based multi-robot collaborative electronic component manufacturing scheduling algorithm module, inter-agent communication latency is less than 10ms, with communication reliability reaching 99.9%. Real-time interaction is achieved through a distributed communication network. During reinforcement learning model training, the experience replay buffer capacity is set to 15,000 data points, and the target network update frequency is once every 120 iterations. In the electronic component manufacturing process optimization algorithm module based on digital twins and virtual simulation, a digital twin model is constructed using professional modeling software such as SolidWorks. The model update frequency is consistent with the sensor data acquisition frequency, ensuring real-time synchronization between the physical entity and the virtual model. By simulating the chip placement stress distribution and circuit board soldering temperature field under different process parameters, potential problems are predicted in advance. Monte Carlo simulation is used to analyze the uncertainty of the prediction results, with the number of simulations set to 1200. In the high-precision chip placement algorithm module based on multi-sensor fusion and real-time correction, the Dempster-Shafer evidence theory uses an improved discount coefficient method to handle conflicting evidence during fusion to improve fusion accuracy. During the Kalman filtering process, the state transfer matrix A and the observation matrix H are determined based on the chip placement dynamics model and sensor measurement characteristics. The process noise covariance Q and the observation noise covariance R are obtained through statistical analysis of experimental data. In the circuit board welding quality optimization algorithm module based on multi-physics field monitoring and adaptive control, when principal component analysis (PCA) is used for dimensionality reduction, the number of principal components is determined by calculating the eigenvalue contribution rate. The fuzzy control rule base is established based on the experience of welding process experts and experimental data, and the fuzzy reasoning adopts the Mamdani reasoning method, and the defuzzification adopts the center of gravity method. In the electronic component rapid detection algorithm module based on multimodal data fusion and deep learning, the deep learning model adopts a structure that combines convolutional neural network (CNN) and recurrent neural network (RNN). CNN extracts the spatial features of electronic components, and RNN processes time series features. The features are further processed through a bidirectional long short-term memory network (Bi-LSTM) to improve the accuracy of defect identification and quantitative analysis. In the multi-robot collaborative electronic component manufacturing scheduling algorithm module based on reinforcement learning, the reward function is based on the manufacturing efficiency improvement ratio α, the product quality improvement ratio β, and the equipment utilization rate improvement ratio γ.

[0048] To construct a quantitative indicator, the reward value R = w1α + w2β + w3γ, and w1 = 0.4 is determined by the hierarchical analysis method.

[0049] , w2=0.3, w3=0.3, to guide the robot to learn the optimal collaborative manufacturing scheduling strategy.

[0050] Example 2

[0051] In a smartphone chip mounting workshop, two similar production lines were selected for comparative testing.

[0052] Traditional placement method group: using traditional visual positioning and fixed placement parameters. The accuracy of the visual sensor is

[0053] ±0.1mm. The placement process relies on a pre-set program, making it difficult to accommodate subtle variations in chip size and circuit board flatness. When placing 1,000 chips, 80 chips were scrapped due to placement position deviation, resulting in an 8% scrap rate. The average chip position deviation after placement was ±0.08mm, exceeding industry standards and impacting the stability of mobile phone performance. Furthermore, due to the frequent manual intervention required for adjustments, the average number of chips placed per hour was 120, resulting in low production efficiency.

[0054] Multi-sensor fusion and real-time correction algorithm: A visual sensor with an accuracy of ±0.01mm, a pressure sensor with an accuracy of ±0.05N, a displacement sensor with an accuracy of ±0.02mm, and a laser ranging sensor with an accuracy of ±0.03mm are deployed on the industrial robot's placement head. The deployment locations were determined through finite element simulation and actual testing. During the placement preparation phase, the sensors collect data at a 10Hz frequency, increasing to 50Hz during the placement execution phase. Let Pv be the chip position data detected by the visual sensor, Fp be the placement pressure data detected by the pressure sensor, Ds be the nozzle displacement data detected by the displacement sensor, and Hl be the circuit board flatness data detected by the laser ranging sensor. After fusion using the Dempster-Shafer evidence theory, the fused data M is generated. Based on this data, a Kalman filter is used to optimize the placement path. In the same 1,000-chip placement task, only 10 chips were rejected, reducing the scrap rate to 1%. The average chip position deviation after placement is kept within ±0.02mm, meeting high-precision requirements. At the same time, because automated real-time correction reduces manual intervention, the average number of chips mounted per hour has increased to 180, and production efficiency has increased by about 50%.

[0055] Implementation of a circuit board welding quality optimization algorithm based on multi-physics field monitoring and adaptive control

[0056] In a computer motherboard welding workshop, two groups of industrial robot welding equipment were also selected for comparison.

[0057] Traditional soldering methods, which rely on preset soldering parameters and fixed paths, are difficult to adapt to varying motherboard materials, thicknesses, and component layouts. During the soldering process of 100 computer motherboards, 15 exhibited solder joint defects such as cold solder joints and short circuits, a 15% defect rate that severely impacted the motherboard's electrical performance and lifespan. The standard deviation of solder joint strength after soldering was ±3 MPa, demonstrating poor consistency.

[0058] Multi-physics field monitoring and adaptive control algorithm group: The welding equipment is equipped with a temperature sensor with an accuracy of ±1°C, a current sensor with an accuracy of ±0.1A, a visual sensor with an accuracy of ±0.02mm, and an ultrasonic sensor with a resolution of ±0.05mm. The welding temperature data detected by the temperature sensor is T, the welding current data detected by the current sensor is I, the solder joint shape data detected by the visual sensor is S, and the internal structure data of the weld part detected by the ultrasonic sensor is V. After dimensionality reduction and fusion using principal component analysis (PCA), the comprehensive data N is obtained. The welding power is adjusted based on fuzzy control. When welding the same number of 100 computer motherboards, the number of motherboards with welding defects was reduced to 3, and the defect rate dropped to 3%. The standard deviation of the solder joint strength after welding was reduced to ±1MPa, significantly improving consistency and effectively enhancing the quality and reliability of motherboard welding.

Claims

1. An electronic component manufacturing system based on information perception and intelligent optimization, characterized in that: include: The chip high-precision placement algorithm module based on multi-sensor fusion and real-time correction obtains multi-source information in the chip placement process by installing multiple sensors on the placement head of the industrial robot, and uses multi-sensor fusion technology and real-time correction algorithm to achieve high-precision chip placement; the chip position data detected by the visual sensor is Pv, the placement pressure data detected by the pressure sensor is Fp, the nozzle displacement data detected by the displacement sensor is Ds, and the circuit board flatness data detected by the laser ranging sensor is Hl. After fusion through the Dempster-Shafer evidence theory, the fused data M is obtained; for the evidence source Si corresponding to each sensor data, its basic probability distribution function is mi, and through orthogonal sum operation m=m1⊕m2⊕…⊕mn to obtain the basic probability distribution m after fusion, and then determine the fusion data M; based on the real-time correction algorithm, the Kalman filter is used to optimize the placement path. Let the state transfer matrix be A and the observation matrix be H, The process noise covariance is Q, the observation noise covariance is R, the initial state estimate is x^0|0, the covariance estimate is P0|0, at time k, first predict: x^k|k-1=Ax^k-1|k-1, Pk|k-1=APk-1|k-1AT+Q; then update: Kk=Pk|k-1HT(HPk|k-1HT+R)-1, x^k|k=x^k|k-1+Kk(zk-Hx^k|k-1), Pk|k=(I-KkH)Pk|k-1, where zk is the observation value corresponding to the fused data M. The Kalman filter process is used to adjust the placement path and other parameters in real time to ensure chip placement accuracy. The circuit board welding quality optimization algorithm module based on multi-physics field monitoring and adaptive control installs multiple sensors on the welding equipment of the industrial robot to obtain the multi-physics field parameters during the welding process. It uses multi-source information fusion technology and adaptive control algorithms to optimize the circuit board welding quality. Let the welding temperature data detected by the temperature sensor be T, the welding current data detected by the current sensor be I, the solder joint shape data detected by the visual sensor be S, and the internal structure data of the welding part detected by the ultrasonic sensor be V. After dimensionality reduction and fusion through principal component analysis (PCA), the comprehensive data N is obtained. First, the original data matrix X = [T, I, S, V] T is centered and the covariance matrix C = n-11XXT is calculated. Then, the eigenvalues ​​λi and eigenvectors ei of the covariance matrix are solved, and the top m principal components with cumulative contributions exceeding 85% are selected to obtain the transformation matrix W = [e1, e2, …, em] and the comprehensive data N = WTX. Based on the adaptive control algorithm, fuzzy control is used to adjust the welding power. The input of the fuzzy controller is assumed to be the comprehensive data N and the expected deviation e of the welding quality and its rate of change Δe. The output is the welding power adjustment ΔPw. The fuzzy set of input and output variables is defined. The fuzzy set of e is {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}. Through the fuzzy rule base, such as the rule "If e is positive large and Δe is positive small, then ΔPw is positive large", the welding power adjustment ΔPw is obtained through fuzzy reasoning and defuzzification. By adjusting the welding parameters in real time, the welding quality is optimized. The electronic component rapid detection algorithm module based on multimodal data fusion and deep learning achieves rapid detection of electronic components by installing multiple detection devices on the electronic component detection production line to obtain multimodal data of electronic components and using multimodal data fusion technology and deep learning algorithms; The reinforcement learning-based multi-robot collaborative electronic component manufacturing scheduling algorithm module treats multiple industrial robots in the electronic component manufacturing workshop as a multi-agent system, defines the state space, action space, and reward function, and uses the reinforcement learning algorithm to achieve efficient collaborative work among multiple robots; The electronic component manufacturing process optimization algorithm module based on digital twin and virtual simulation establishes a digital twin model of the electronic component manufacturing system and uses virtual simulation to simulate and optimize the manufacturing process of electronic components, thereby achieving optimization and improvement of the electronic component manufacturing process.

2. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the high-precision chip placement algorithm module based on multi-sensor fusion and real-time correction, the accuracy of the visual sensor is ±0.01mm, the accuracy of the pressure sensor is ±0.05N, the accuracy of the displacement sensor is ±0.02mm, and the accuracy of the laser ranging sensor is ±0.03mm. The sensor deployment position is determined by combining finite element simulation and actual testing to ensure accurate data collection. The sensor acquisition frequency is 10Hz in the placement preparation stage, increased to 50Hz in the placement execution stage, and dynamically adjusted according to the placement process.

3. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the circuit board welding quality optimization algorithm module based on multi-physical field monitoring and adaptive control, the temperature sensor accuracy is ±1°C, the current sensor accuracy is ±0.1A, the visual sensor accuracy is ±0.02mm, and the ultrasonic sensor resolution is ±0.05mm. The sensors work together to provide high-frequency and precise data for the adaptive control algorithm, and data transmission adopts a low-latency, high-bandwidth industrial Ethernet protocol.

4. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the electronic component rapid detection algorithm module based on multimodal data fusion and deep learning, the resolution of the visual inspection equipment is 0.01mm, the penetration depth resolution of the X-ray inspection equipment is 0.05mm, and the temperature resolution of the infrared inspection equipment is ±0.2°C. Multimodal data fusion adopts a fusion network based on deep learning. The network structure contains 3 convolution layers, and the convolution kernel sizes are 3×3, 5×5, and 3×3 respectively. The network is trained with a large amount of electronic component defect data (more than 8,000 groups) and the number of training times is 2,500 to improve the defect recognition capability.

5. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the reinforcement learning-based multi-robot collaborative electronic component manufacturing scheduling algorithm module, the communication delay between intelligent agents is less than 10ms, the communication reliability reaches 99.9%, and real-time interaction is achieved through a distributed communication network. When training the reinforcement learning model, the experience replay buffer capacity is set to 15,000 data points, and the target network update frequency is once every 120 iterations.

6. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the electronic component manufacturing process optimization algorithm module based on digital twins and virtual simulation, professional modeling software such as SolidWorks is used to build a digital twin model. The model update frequency is consistent with the sensor data acquisition frequency to ensure real-time synchronization between the physical entity and the virtual model. By simulating the chip mounting stress distribution and circuit board welding temperature field under different process parameters, potential problems are predicted in advance, and Monte Carlo simulation is used to perform uncertainty analysis on the prediction results. The number of simulations is set to 1200.

7. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the high-precision chip placement algorithm module based on multi-sensor fusion and real-time correction, the Dempster-Shafer evidence theory uses an improved discount coefficient method to process conflicting evidence during fusion to improve fusion accuracy. During the Kalman filtering process, the state transition matrix A and the observation matrix H are determined based on the chip placement dynamics model and sensor measurement characteristics. The process noise covariance Q and the observation noise covariance R are obtained through statistical analysis of experimental data.

8. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the circuit board welding quality optimization algorithm module based on multi-physical field monitoring and adaptive control, when principal component analysis (PCA) is used for dimensionality reduction, the number of principal components is determined by calculating the eigenvalue contribution rate. The fuzzy control rule base is established based on the experience of welding process experts and experimental data. The fuzzy reasoning adopts the Mamdani reasoning method, and the defuzzification adopts the center of gravity method.

9. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the electronic component rapid detection algorithm module based on multimodal data fusion and deep learning, the deep learning model adopts a structure that combines convolutional neural networks (CNN) and recurrent neural networks (RNN). CNN extracts spatial features of electronic components, and RNN processes time series features. The features are further processed through a bidirectional long short-term memory network (Bi-LSTM) to improve the accuracy of defect identification and quantitative analysis.

10. The key technology for manufacturing electronic components of industrial robots based on multi-source information perception and intelligent optimization according to claim 1 is characterized in that: In the multi-robot collaborative electronic component manufacturing scheduling algorithm module based on reinforcement learning, the reward function is constructed with the manufacturing efficiency improvement ratio α, product quality improvement ratio β, and equipment utilization rate improvement ratio γ as quantitative indicators, that is, the reward value R = w1α + w2β + w3γ. The hierarchical analysis method determines w1 = 0.

4. w2=0.3, w3=0.3 to guide the robot to learn the optimal collaborative manufacturing scheduling strategy.

Images